A Method Of Mechanical Fault Features Based On Morphological Filtering And Blind Signal Processing

Types of Mechanical equipment faults are becoming more complicated since the equipment turns to maximization, automation and integration. This kind of fault would be resulted in disasters and negative social impacts. Therefore, condition monitoring and fault diagnosis for machine equipment have a great significance. Fault feature extraction is the key for machinery fault diagnosis. As known to all, rolling bearings are widely used in variety of rotating machinery, so further study on the fault signal and fault feature extraction methods of rolling bearing has great practical significance.In this thesis, the morphological filtering and blind signal processing are used in rolling bearing fault feature extraction, particularly, it has researched on algorithms of blind deconvolution and non-negative matrix factorization which are used in mechanical fault feature extraction. Mathematical morphology filtering was applied in reducing the background noise in fault signal because of its nonlinear, non-stationary characteristics. An amount of morphological filters have been studied and many simulation experiments have been made to get the filtering more effective. On the basis of blind deconvolution algorithm based on independent component analysis and cluster analysis, a time-domain blind deconvolution algorithm including mathematical morphology filtering, improved KL distance algorithm, and orthogonal matching pursuit algorithm has been put forward. Finally the algorithm was verified by simulation test. This thesis also studied three initialization algorithm of non-negative matrix factorization (the initialization algorithm based on multiple random distribution, the initialization algorithm based on SVD, the initialization algorithm based on FCM), and two non-negative decomposition algorithm (the non-negative decomposition algorithm based on multiplicative iterative, the non-negative decomposition algorithm based on alternating least squares method).Finally, the data acquisition experiments of the rolling bearing fault signal were made on the QPZZ-Ⅱ rotating machinery fault simulation test bench. The experiment results of bearing compound fault extraction in real working-environment demonstrate the accuracy and reliability of the proposed algorithm.